MIT researchers are figuring out the tougher part of self-driving: Building software and adapting sensors so vehicles can drive on country roads that haven’t yet been 3D-mapped and repeatedly test-driven. More than a third of US roads are unpaved, and others are not lit or lack well-marked road edges. These are also the roads that have the highest fatality rates (per mile driven).
MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) has developed MapLite, described as a “framework that allows self-driving cars to drive on roads they’ve never been on before without 3D maps.”
MapLite — a nice play on words from when overhead lights were for reading paper maps — combines basic GPS map data, basic as in what you’d find on Google Maps, with sensors that keep close watch on road conditions. The big issues are detecting unmarked road edges when the road is dark, and the gravel or dirt beyond the road is dark, too (image above).
Working on unpaved roads in Devens, Massachusetts, and collaborating with Toyota Research, which supplied a Prius as the moving test bed, the testers have been able to reliably detect the road edges 100 feet ahead. At 30 mph, that would give the car 2.2 seconds to safely stop, or 1.5 seconds at a more adventurous 45 mph. At a “hold my beer and watch this” 60 mph, there might be problems. That’s our observation, not MIT’s.
How far along is the project? It’s an important first step, says Daniela Rus, director of MIT’s Computer Science and Artificial Intelligence Laboratory. “The need for dense 3-D maps limits the places where self-driving cars can operate.”
Able to Handle Roads the Big Map Companies Don’t Map
According to MIT CSAIL grad student Teddy Ort, “The reason this kind of ‘map-less’ [or basic mps] approach hasn’t really been done before is because it is generally much harder to reach the same accuracy and reliability as with detailed maps. A system like [MapLite] that can navigate just with on-board sensors shows the potential of self-driving cars being able to actually handle roads beyond the small number that tech companies have mapped.”
If a car has detailed digital maps to work from, then it can turn the task of tracking deviations from the map. That could include street barricades, stopped vehicles, pedestrians crossing, and bicyclists wavering along the side of the road.
What MapLite Can and Can’t Do
MapLite uses lidar and other sensors for navigation. GPS data is there only to obtain an estimate of the car’s location. MapLite sets a final destination, plus researches a “local navigation goal,” or what’s within the current view from the car. The perception sensors create a path to that point, using lidar to determine — estimate — the road’s edges.
One assumption helping MapLite: It presumes the road will be flatter than the surrounding areas. That helps with edge detection. The MIT researchers also developed models that are “parameterized,” meaning they describe situations that are somewhat similar. One model might be broad enough to determine what to do at intersections, with another for a specific type of road.
Limitations remain. According to the MapLite team, the biggest challenge is mountain roads, because the system has trouble dealing with dramatic elevation changes.
How to Build a Face Mask Detector With a Jetson Nano 2GB and AlwaysAI
Nvidia continues to make AI at the edge more affordable and easier to deploy. So instead of simply running through the benchmarks to review the new Jetson Nano 2GB, I decided to tackle the DIY project of building my own face mask detector.
This Is a Bad Time to Build a High-End Gaming PC
We're not going to say it's the worst time to build a high-end gaming PC, but if you'll need to get lucky with some orders if you want to pull it off this year.
Samsung, Stanford Built a 10,000 PPI Display That Could Revolutionize VR, AR
Ask anyone who has spent more than a few minutes inside a VR headset, and they'll mention the screen door effect. This could eliminate it for good.